How Ontario employers word their AI hiring disclosures: a study of 2,141 job postings

Since 1 January 2026, Ontario employers with 25 or more employees have had to say in a public job posting if they use artificial intelligence to screen, assess or select applicants. Indeed has shown that AI mentions in Ontario postings tripled after the rule took effect. Nobody has published how employers actually word the statement. So we collected every Ontario posting on Workable’s public job board on 5 November 2026 and read every AI statement in them.

What we found, across 2,141 postings from 250 employers:

• 67 employers (27%) had an AI statement in at least one posting. 52 said they use or may use AI, and 15 said they don’t.
• Most statements say little about what the AI does. Of the 52 employers that disclosed AI use, 29 (56%) gave only a general statement, such as “AI technology supports the application screening process.”
• A third hedge. 17 of the 52 (33%) said only that they may use AI.
• Fewer than a third mention people. 16 of the 52 (31%) said a person reviews applications or makes the decisions.
• “No AI” statements echo the law; “yes” statements don’t. 10 of the 15 employers saying they don’t use AI repeated the Act’s phrase “screen, assess or select”. Only 3 of the 52 disclosing use did.
• Statements come and go between postings. 21 of the 170 employers with more than one 2026 posting included an AI statement on some of their postings but not others.
• Employers that followed the other new rules usually addressed AI too. Of the 49 employers that said whether a job was an existing vacancy, another 2026 requirement, 40 (82%) also had an AI statement. Of the 201 that didn’t, 27 (13%) did.

An important caution: a posting without an AI statement is not necessarily breaking the rule. The rule only applies to employers with 25 or more employees that use AI to screen, assess or select applicants, and the postings don’t tell us either of those things. We don’t describe any posting or employer as non-compliant, and we don’t name employers.

The method, limits and an anonymised dataset are at the end.

Why we did this

Indeed’s Hiring Lab tracks how many postings mention AI. In Ontario, the share rose from 9% in October 2025 to 28% in May 2026, and Indeed attributes the jump to the disclosure rule (Woessner and Bernard, 2026). By 30 September, 19% of all Canadian postings mentioned AI, against 7.3% in the U.S. (Indeed Hiring Lab, 2026).

Those figures count mentions. They don’t show what the statements say, which is what applicants actually read, and what employers writing their own statement want to know. The Ministry of Labour’s guidance says only that it is “enough for the employer to state that artificial intelligence is used to screen, assess or select applicants” (Ontario ESA guide). This study looks at what employers did with that freedom.

The findings in detail

Counts are per employer: an employer counts once if any of its postings has the feature. We lead with employers because a few post a lot. One employer had 302 postings and another 148, while the median employer had two.

Of 250 employers	Employers	Share
Any AI statement	67	27%
Says it uses AI, in at least one posting	35	14%
Says only that it may use AI	17	7%
Says it doesn’t use AI	15	6%
Said whether the job is an existing vacancy (any wording)	49	20%

Of the 52 employers disclosing AI use	Employers	Share
Says something specific about what the AI does	23	44%
Mentions human review or human decisions	16	31%
Both specific and mentions human review	11	21%
Attributes the AI to a vendor tool or applicant tracking system	12	23%
Uses the Act’s phrase “screen, assess or select”	3	6%

By posting rather than employer, 587 of 2,141 postings (27%) had an AI statement. Leaving out the largest employer, whose 302 postings had none, it was 587 of 1,839 (32%).

Most statements are general

The most common type says AI is involved in screening and stops there:

Please note that AI technology supports the application screening process.

As part of our hiring process, we use AI-powered tools to assist in screening resumes.

That second sentence appeared word for word at four different employers. No sentence appeared at more than four, so the statements on this platform aren’t a vendor’s standard text. Employers are writing their own, or copying from the same few sources.

Specific statements say which task the AI does, at which stage, or with which tool:

Our hiring process may include AI screening for keywords and minimum qualifications.

We use AI-assisted tools to support our team with tasks like organizing applications and highlighting job-related qualifications.

Tasks named included sorting, matching, ranking, parsing, keyword screening, summarising, anonymising applications, analysing interview notes and scheduling. One employer offered a list: “application review support, candidate research, interview preparation, scheduling support, and workflow administration.”

“May” is common

A third of the employers disclosing AI said only that they might use it:

AI technology may be used in selecting candidates for interviews.

Our organization may use Artificial Intelligence (AI) enabled tools to support our recruitment initiatives.

“May” makes sense when it really varies by role. When it doesn’t, it tells the applicant less than a plain statement would. The guidance asks for a statement disclosing the employer’s use of AI, and doesn’t say how hedged wording is treated.

Human review is mentioned less often than you might expect

Indeed observed that many postings with AI disclosures “also emphasize that human decision-makers have the final say”. In this sample, 16 of the 52 employers disclosing AI use did. The strongest examples said who reviews and when:

AI does not make the final decision: our CFO reviews every application, with the benefit of that AI analysis, before any candidate is ruled out.

Our applicant tracking system uses AI-enabled tools to help sort applications based on job-related criteria, but members of our hiring team personally review all applications and make all hiring decisions.

“No AI” statements copy the law; “yes” statements paraphrase

Employers that don’t use AI tended to answer in the Act’s words: “Artificial intelligence is not used to screen, assess, or select applicants for this role.” Employers that do use AI almost never did. They wrote “supports”, “assists”, “helps sort”, “streamline”. One employer used the guidance’s wording almost exactly: “Artificial intelligence is used to screen, assess or select applicants.”

One “no AI” statement made a useful distinction: “We don’t use AI to evaluate applications, though you may be automatically screened out if you don’t meet baseline requirements.” An automatic knockout rule isn’t necessarily AI. Our guide to the rule covers where that line probably falls.

The statement isn’t always on every posting

Of the 170 employers with more than one posting from 2026, 21 had an AI statement on some postings and not on others. In some cases it was missing from one or two of dozens, which looks like a template that didn’t reach every posting. In others, most postings lacked it. Employers sometimes use AI for some roles and not others, so a difference between postings isn’t proof of an error. But if the same tools screen every role, the statement belongs on every posting.

Consent and opt-outs

Four employers went beyond disclosure and asked for consent: “By clicking ‘Apply,’ you acknowledge and consent to the use of AI in our recruitment workflow.” One offered a way out: “If you prefer that your CV not be processed using AI, please let us know,” adding that applicants who opt out are still considered. The Act asks for neither.

What this means if you’re writing yours

These are our recommendations, based on the findings above and the Ministry’s guidance. They are not legal advice.

1. Say what the AI does. A sentence naming the task (“sorts applications against the job requirements”) tells applicants more than “supports our process”, and costs nothing extra.
2. Drop “may” if you always use it.
3. Mention human review only if it’s true for this job. If any tool can set applicants aside without a person looking, don’t say a person makes every decision.
4. Put the same statement on every posting that uses the same process, including job boards and recruiters’ versions. Check your templates.
5. If you don’t use AI, saying so is optional. The 15 employers that did used the Act’s wording, which is a clear way to do it.

Our guide to Ontario’s AI disclosure rule covers who is covered, what counts as AI and sample wording.

Method

Source. Workable, an applicant tracking system, runs a public job board (jobs.workable.com) listing open postings from employers that use it. On 5 November 2026 we requested every result for the location “Ontario, Canada” through the board’s public search endpoint, at about one request per second. That returned 5,095 postings.

Ontario filter. The search also returned postings in other provinces. We kept postings whose province field was Ontario, plus 37 that had no province or a city or region in that field (such as “Waterloo Regional Municipality”) and whose city appears as an Ontario city elsewhere in the data. That left 2,141 postings from 250 employers. 2,034 (95%) were created in 2026; the earliest was from July 2021. 1,806 were in English, 331 in French and 4 had no language set.

Finding statements. For each posting we took the description, requirements and benefits text and split it into sentences. We listed every distinct sentence that paired an AI term (“artificial intelligence”, “AI”, “machine learning”, “automated tools”, “algorithm” and French equivalents) with a hiring term (“applicant”, “candidate”, “screen”, “assess”, “select”, “recruit”, “resume” and French equivalents), and every sentence mentioning a vacancy: 300 distinct sentences.

Coding. Each of the 300 sentences was read and coded against written rules: whether it says the employer uses AI, may use AI or doesn’t; whether it names a task, stage or tool (specific) or not (general); whether it mentions human review or decisions; whether it attributes the AI to a vendor tool or platform; whether it uses all three of the Act’s verbs; and whether it says if the job is an existing vacancy. Sentences that only add to a statement (“AI does not make final hiring decisions”) count toward its features. 115 of the 300 were not disclosures, mostly AI skills in job descriptions. The column definitions under how to use the data give the rules.

Checking for misses. Pattern matching can miss a statement worded unusually. After coding, we read every posting with no AI statement whose text paired an AI term with a hiring word, and every short sentence pairing an AI term with “use”, “tool” or “disclose”. Neither check found a missed disclosure.

Who did the coding. The coding was done with AI assistance, as described on our editorial page, and reviewed by a person on our team. It was not repeated by a second, independent coder, so we haven’t measured agreement between coders. The two judgement calls most likely to differ are “specific” versus “general” and whether a sentence mentions human review; the counts of AI and vacancy statements are much less sensitive to judgement.

Limits

• One platform. Workable’s customers are not a random sample of Ontario employers, and these are the postings on one job board on one day. The results describe that sample, not all Ontario postings.
• We can’t see who is covered. The postings don’t show employer size, so we can’t separate employers under the 25-employee threshold. They also don’t show whether an employer uses AI. A missing statement can mean the rule doesn’t apply.
• Concentrated. A few employers account for many postings, which is why we report by employer. The posting-level figure moves from 27% to 32% when the largest employer is left out.
• One snapshot. The scripts and coding rules are written so the study can be repeated on a later date and compared.
• A different measure from Indeed’s. Indeed counts postings on its own site that mention AI-related terms, including AI skills. Our figures count only hiring disclosures on Workable’s board, so the two shouldn’t be compared directly.

How to use the data

The anonymised dataset has one row per posting, with codes but no employer names, job titles or posting text: ontario-ai-disclosure-study-2026-11.csv (2,141 rows). Employers are numbered (E0000 onwards) so postings from the same employer can be grouped. The columns are:

Column	Meaning
created_month	Month the posting was created
since_2026	Created on or after 1 January 2026
language	en or fr
workplace	On site, hybrid or remote, as listed
ai_statement	uses_ai, may_use_ai, no_ai or blank (none)
detail	specific (names a task, stage or tool) or minimal, for statements of use
human_review	Mentions that a person reviews applications or makes decisions
vendor_tool	Attributes the AI to a vendor tool, applicant tracking system or platform
statute_wording	Uses all three of “screen”, “assess” and “select”
vacancy_statement	Says whether the posting is for an existing, current or new vacancy
notes	consent, opt_out, privacy, training_data or knockout

You’re welcome to use it with a link to this page. To cite the study: SortTheStack (2026), “How Ontario employers word their AI hiring disclosures: a study of 2,141 job postings”, collected 5 November 2026.

Where SortTheStack fits

SortTheStack uses AI to assess applications against the requirements you write, so jobs screened with it need the statement where the Ontario rule applies. Postings on its careers page include an AI statement automatically, on every job. If you post the same job elsewhere, add the statement there too. See how the screening works.

Sources

• Ontario Ministry of Labour, Immigration, Training and Skills Development, Your guide to the Employment Standards Act: Requirements related to publicly advertised job postings (updated 24 February 2026), accessed 5 November 2026.
• Woessner, A., & Bernard, B. (2026, July 8). Ontario job postings highlight the widespread use of AI in hiring processes. Indeed Hiring Lab Canada.
• Indeed Hiring Lab, Q3 2026 Canadian labour market update (27 October 2026), accessed 5 November 2026.
• Workable, public job board, jobs.workable.com, postings collected 5 November 2026.